CN109242797A - Image de-noising method, system and the medium merged based on homogeneous and heterogeneous areas - Google Patents
Image de-noising method, system and the medium merged based on homogeneous and heterogeneous areas Download PDFInfo
- Publication number
- CN109242797A CN109242797A CN201811061739.8A CN201811061739A CN109242797A CN 109242797 A CN109242797 A CN 109242797A CN 201811061739 A CN201811061739 A CN 201811061739A CN 109242797 A CN109242797 A CN 109242797A
- Authority
- CN
- China
- Prior art keywords
- sub
- block
- image
- denoising
- heterogeneous areas
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 52
- 230000004927 fusion Effects 0.000 claims abstract description 12
- 238000013507 mapping Methods 0.000 description 12
- 230000000694 effects Effects 0.000 description 7
- 238000001914 filtration Methods 0.000 description 7
- 238000000513 principal component analysis Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000011524 similarity measure Methods 0.000 description 4
- 230000009466 transformation Effects 0.000 description 4
- 230000008859 change Effects 0.000 description 2
- 230000000295 complement effect Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 235000007926 Craterellus fallax Nutrition 0.000 description 1
- 240000007175 Datura inoxia Species 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000005315 distribution function Methods 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 239000012085 test solution Substances 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The invention discloses image de-noising method, system and media based on homogeneous and heterogeneous areas fusion, include: step (1): setting the sliding step of window, original image is divided into several sub-blocks according to being sized for window, calculates the weight coefficient of each sub-block；Step (2): all sub-blocks divided to original image carry out denoising using LPG-PCA algorithm；To all sub-blocks that original image divides, denoising is carried out using three-dimensional Block- matching BM3D algorithm；Step (3): according to weight coefficient compared with given threshold, each sub-block is classified as homogenous area or heterogeneous areas；According to the area classification and weight coefficient of each sub-block, the sub-block after two kinds of algorithms are denoised carries out corresponding fusion, and obtaining fused image is final image.
Description
Technical field
The present invention relates to image de-noising method, system and media based on homogeneous and heterogeneous areas fusion.
Background technique
With the rapid development of image technique, for image in medical imaging, pattern-recognition etc. achieves extensive use.But
It is that image is being formed, in transmission process, inevitably will receive the interference of noise.Noise in image is often handed over signal
It is woven in together, makes the details of image itself, such as: profile and border, lines etc. become blurred.Therefore noisy acoustic image is carried out
Denoising is very necessary, convenient for higher level image analysis and understanding.And how both the noise occurred in image to be carried out
It is reasonable to inhibit, and the information of the different needs of removal, and the useful information in image can be made to be strengthened, consequently facilitating target
Differentiation or intercept, are the problem of image denoising should be studied mainly.
We are handled according to the difference of the numerical characteristic of noise and its gray value and ambient signals gray value in practice,
Denoising process can both be completed in image space domain, can also be completed in image transform domain.Image space domain denoising is original
Data operation is carried out on image, and directly the gray value of pixel is handled.Currently, image space domain denoising method has mean value filter
Wave, median filtering and low-pass filtering etc., these methods have a common feature, i.e., each pixel in image is by with same
Kind mode is handled, and the characteristic without considering each pixel itself is highly effective in terms of removing additional random noise, but is going
While except noise, also occur image more serious fuzzy, especially the edge of image and at details, obscure compared with
It is serious.Another kind of very effective image de-noising method is the denoising based on image transform domain, and basic thought is: right first
Noisy image carries out certain transformation, it is changed to transform domain from transform of spatial domain, then carries out again to the transformation coefficient in transform domain
Processing, carries out inverse transformation later, final to realize effectively denoising by noisy image from transform domain reconvert to original spatial domain.
However, due to the difference of characteristics of image, the strong changeability of the physical attribute and data of image between images has
Image texture structure is compared with horn of plenty, and some image homogeneous part is more, and current Denoising Algorithm, either based on spatial domain
Or transform domain, it is all based on centainly simplified iconic model, their denoising performance is only more prominent at single aspect, therefore
Single denoising model can not be applied to different characteristic image denoising and obtain excellent effect.So to image by feature point
Targetedly denoising is of great significance after class.Occur many advanced Denoising Algorithms in recent years, such as three-dimensional Block- matching is calculated
Method (Block Matching 3D, BM3D), non local homomorphism sparse coding (Non-local Sparse Code, LSSC) and non-
The quick self-adapted SAR image Denoising Algorithm (Fast adaptive nonlocal SAR despeckling, FANS) in part, office
Principal Component Analysis (the Principal Component Analysis With Local Pixel of portion's block of pixels grouping
Grouping, LPG-PCA), non local average (Non-local Means, NLM) Denoising Algorithm, the dictionary for rarefaction representation
Learning algorithm (K-Means-Singular Value Decomposition, K-SVD) etc., the denoising ability that they have
Strong but weaker to the protection of the details of image, some noise removal capabilities are weak but preferably protect the texture of image.Currently, without one
The kind existing stronger denoising ability of method can preferably protect original image details again, while not generate pseudo- shadow information.
Summary of the invention
In order to solve the deficiencies in the prior art, the present invention provides the image denoising sides based on homogeneous and heterogeneous areas fusion
Method, system and medium；The value of algorithm denoising performance index peak signal-to-noise ratio and structural similarity of the invention more individually denoises
Algorithm is promoted, and visual effect and details protection are also better than single Denoising Algorithm.
In order to solve the above-mentioned technical problem, the present invention adopts the following technical scheme:
As the first aspect of the present invention, the image de-noising method based on homogeneous and heterogeneous areas fusion is proposed；
The image de-noising method merged based on homogeneous and heterogeneous areas, comprising:
Step (1): setting the sliding step of window, and original image is divided into several height according to being sized for window
Block calculates the weight coefficient of each sub-block；
Step (2): all sub-blocks divided to original image carry out denoising using LPG-PCA algorithm；
To all sub-blocks that original image divides, denoising is carried out using three-dimensional Block- matching BM3D algorithm；
Step (3): according to weight coefficient compared with given threshold, each sub-block is classified as homogenous area or heterogeneous area
Domain；According to the area classification and weight coefficient of each sub-block, the sub-block after two kinds of algorithms are denoised carries out corresponding fusion, is melted
Image after conjunction is final image.
Further, the homogenous area refers to: the gray value of all pixels point is in a setting range in the region
It is interior.
Further, the heterogeneous areas refers to: the region other than homogenous area.
Further, in the step (1), the step of calculating the weight coefficient of each sub-block are as follows:
Step (101): the Generalized Likelihood Ratio λ of j-th of sub-block is calculatedj(x) are as follows:
Wherein, G expression is the geometric mean of j-th of sub-block, and A expression is the arithmetic equal value of j-th of sub-block,
N indicates the total number of pixel in j-th of sub-block, xiIndicate i-th of picture
The pixel value of element；
Step (102): according to Generalized Likelihood Ratio λj(x) weights omega (λ of j-th of sub-block is calculatedj):
Wherein, parameter lambda0The median of all sub-block Generalized Likelihood Ratio λ (x) is taken, slope is setting value.
Further, the step of step (3) are as follows:
To j-th of sub-block of image, first determine whether that it belongs to homogenous area or heterogeneous areas；
If weight coefficient ω (λj) being more than or equal to 0.5, then sub-block belongs to heterogeneous areas；For belonging to the son of heterogeneous areas
Block will carry out the grey scale pixel value of j-th of sub-block after denoising using three-dimensional Block- matching BM3D algorithm multiplied by weights omega
(λj) after, the grey scale pixel value with j-th of sub-block for carrying out denoising using LPG-PCA algorithm is multiplied by (1- ω (λj)) after ask
With obtain the grey scale pixel value of j-th of sub-block of fused image；At this moment, three-dimensional Block- matching BM3D algorithm is to the sub-block tribute
It offers slightly larger, can preferably protect the detailed information of heterogeneous areas.
If weight coefficient ω (λj) less than 0.5, then sub-block belongs to homogenous area；For belonging to the sub-block of homogenous area,
The grey scale pixel value of j-th of sub-block of denoising will be carried out using three-dimensional Block- matching BM3D algorithm multiplied by weights omega (λj) after,
Grey scale pixel value with j-th of sub-block for carrying out denoising using LPG-PCA algorithm is multiplied by (1- ω (λj)) sum afterwards, it obtains
The grey scale pixel value of j-th of sub-block of fused image.At this moment, LPG-PCA algorithm is contributed the sub-block slightly larger, can be preferably
Smooth homogenous area, achievees the purpose that denoising.
Obtain the grey scale pixel value of all sub-blocks of fused image in turn to get the figure after original image denoising is arrived
Picture.
Explanation of nouns:
LPG-PCA: Principal Component Analysis (the Principal Component Analysis of local pixel block grouping
With Local Pixel Grouping, LPG-PCA).The algorithm by training set pixel to be processed and its neighborhood be expressed as
Sub-block vector, is grouped sub-block to obtain sample matrix using block similarity measurement, carries out centralization to sample matrix, then
Denoising is carried out using principal component analysis.
BM3D: three-dimensional block matching algorithm (Block Matching 3D, BM3D).The algorithm by non-local filtering method with
Wavelet shrinkage and Wiener filtering combine.First with block matching algorithm by each sub-block to be processed and its similar sub-block group
At three-dimensional storehouse, then it is filtered in 3 D wavelet transformation domain using hard -threshold, filtered sub-block is put back in original image
Position.Filtered image is further processed using block matching algorithm and Wiener filtering, obtains denoising image to the end.
Further, by adding to the image after LPG-PCA and BM3D denoising according to homogenous area and heterogeneous areas
Power fusion, it is ensured that be effectively reduced noise in homogenous area, and avoid generating artificial artifact；It is preferably protected in heterogeneous areas
The texture and details of image.
As a second aspect of the invention, the image denoising system based on homogeneous and heterogeneous areas fusion is proposed；
The image denoising system merged based on homogeneous and heterogeneous areas, comprising: memory, processor and be stored in storage
The computer instruction run on device and on a processor, when the computer instruction is run by processor, completes any of the above-described side
Step described in method.
As the third aspect of the present invention, a kind of computer readable storage medium is proposed；
A kind of computer readable storage medium, is stored thereon with computer instruction, and the computer instruction is transported by processor
When row, step described in any of the above-described method is completed.
Compared with prior art, the beneficial effects of the present invention are:
It proposes a kind of soft classification policy, carries out good compromise in protection details and two aspect of denoising.The algorithm is based on working as
The existing advanced denoising basic tool of first two, key are two kinds of excellent Denoising Algorithms with complementary characteristic of selection,
Emphasis is the design of combinator.The present invention is by the way of soft-combine, in conjunction with the weight coefficient linear convergent rate changed from 0 to 1
The linear combination of two algorithms optimizes the denoising effect of single Denoising Algorithm at present.The denoising and classification of image are currently to grind
The hot issue studied carefully, there is important research significance.Therefore the Image denoising algorithm based on soft classification of invention research exists
The normal image denoising application of non local method has far reaching significance.
Detailed description of the invention
The accompanying drawings constituting a part of this application is used to provide further understanding of the present application, and the application's shows
Meaning property embodiment and its explanation are not constituted an undue limitation on the present application for explaining the application.
Fig. 1 is flow chart of the invention；
Fig. 2 (a)-Fig. 2 (h) is to choose the comparison diagram that different windows carries out soft classification to Lena image；
Fig. 3 (a)-Fig. 3 (h) is the comparison diagram that distinct methods denoising is used to noise-containing Lena image.
Specific embodiment
It is noted that following detailed description is all illustrative, it is intended to provide further instruction to the application.Unless another
It indicates, all technical and scientific terms used herein has usual with the application person of an ordinary skill in the technical field
The identical meanings of understanding.
It should be noted that term used herein above is merely to describe specific embodiment, and be not intended to restricted root
According to the illustrative embodiments of the application.As used herein, unless the context clearly indicates otherwise, otherwise singular
Also it is intended to include plural form, additionally, it should be understood that, when in the present specification using term "comprising" and/or " packet
Include " when, indicate existing characteristics, step, operation, device, component and/or their combination.
As shown in Figure 1, based on the image de-noising method that homogeneous and heterogeneous areas merge, the specific steps are that:
Step 1: image is subjected to soft classification according to texture, is divided into homogenous area and heterogeneous areas；
Step 2: currently advanced Image denoising algorithm is utilized, different methods is respectively adopted and is denoised；
Step 3: effectively merging the denoising image that distinct methods obtain, to significantly reduce not same district
The noise level in domain, while protecting the texture and details of image.
The specific steps of the step 1 are as follows:
Step (1-1): for image is classified according to homogenous area and heterogeneous areas, using the arithmetic equal value of gray-value image
The ratio of A and geometric mean G is as the statistical information for distinguishing the two.By a suitable big window centered on object pixel
Mouth calculates, to reduce the damaging influence of noise.To provide the intensive region of full resolution, we enable the window exist
Entire image sliding.
Step (1-2-1): the size for increasing the window of calculating A/G ratio can reduce the variance of estimated value, but will lead to
Estimate mistake.This is because the pixel into window has heterogeneity after increasing window.In order to solve this problem, I
Take a hint from the method for non local denoising.In order to comprising more pixels, using a relatively large region of search,
Those are selected to be more likely to and the consistent pixel of object pixel using block similarity measure.Specifically, we biggish search at one
It is found and K most like sub-block of sub-block to be processed within the scope of rope using block similarity measure.This K sub-block constitutes three-dimensional heap
Stack.Then three-dimensional storehouse is averaged along the third dimension, A/G statistics is calculated by obtained average sub-block.This is equivalent to
Viewization step more than one is used in terms of improving the quality of data, although this is the strategy of a rather rough, reliable
Property aspect have significant improvement.
Step (1-2-2): under the hypothesis of some simplification, A/G is Generalized Likelihood Ratio (Generalized Likelihood
Ratio, GLR) test solution, therefore we pay close attention to the ratio of A and G.For containing N number of pixel centered on giving object pixel
Image block, it is proposed that two hypothesis:
H0Homogenous area: the signal amplitude in image block is equal；
H1Heterogeneous areas: the signal amplitude in image block is unequal.
Corresponding GLR are as follows:
Wherein, x is pixel value, Xi|uiIt obeysDistribution.It willGeneration
Enter Λ (x), and seek sup, we finally obtain Generalized Likelihood Ratio (GLR) statistical presentation formula, logarithmic form are as follows:
Wherein, G expression is the geometric mean of j-th of sub-block, and A expression is the arithmetic equal value of j-th of sub-block,
N indicates the total number of pixel in j-th of sub-block, xiIndicate i-th of picture
The pixel value of element；
Step (1-2-3): it by establishing three-dimensional storehouse in step (1-2-1), is similar to virtual visualization and handles, improve
The quality of weight mapping graph.Relatively small estimation window is used, reliable statistical result can be also provided, to ensure that height
Resolution ratio.However, if using fixed size sub-block window, will lead to window edge occur it is discontinuous.This phenomenon be due to
Each pixel belongs to the estimation window of its many adjacent pixels, affects their heterogeneity index.For example, uniformly carrying on the back
The pixel that very noisy on scape will lead in entire window is marked as heterogeneous areas.Finally, neighborhood should belong to homogeneous area
The pixel in domain is taken as heterogeneous areas to denoise, and generates artificial artifact in the output.In order to solve this problem, we use
One simple heuristic strategies, i.e. weight of the calculating based on sub-block size, wherein may include biggish weight and lesser
Then weight is averaged to different size of sub-block.Some unreasonable estimations can be reduced by doing so, and avoid manually filtering
The appearance of artifact.
Step (1-3): theoretically, for reasonable categorised decision, we only need to be arranged a threshold value.Although not yet
There is λ (x) probability-distribution function represented, but is distributed by gamma it is found that we recognize at approximation when N is the larger value (N > 10)
N and L are only dependent upon for its shape and scale parameter.Therefore, a desired false-alarm probability can be set, and be any of N and L
Exploitation analyzes corresponding thresholding.However, there is no the real images for really solving us to classify for these theoretic results
Problem, because all categorised decisions all become unreliable when noise is stronger, unless a large amount of sample is included in survey by us
Examination.Therefore, crucial problem is the size N (comprising number of pixels) of sampling window, this is related to whether we can obtain reliably
Resolution ratio.In a wicket, even if decision statistics is also very difficult in homogenous area；On the other hand, if increased
N, analysis window can be easy to comprising pixel target of different nature, so as to cause more frequent error result.Due to geometry
Average value is influenced to will lead to serious error by discrete sample strongly.Therefore N value is excessive or too small both of which can not
Obtain good classification denoising effect.For these problems, basic classification method is supplemented:
1) soft classification；
2) virtual more viewization processing；
3) multi-resolution hierarchy.
These three problems will briefly be illustrated below:
Soft classification
Soft classification is the core of our proposed Denoising Algorithms, we have two in image homogeneous with heterogeneous areas
The Denoising Algorithm of complementary characteristic is effectively combined.The characteristic of Denoising Algorithm depends on the property of image, and therefore, we select
Most can the parameter of representative image property adjust weight coefficient.We do not use hard -threshold to handle, but smooth by one
Non-linear logistic function calculates weight coefficient, it with characteristics of image change and between [0,1] value.
Use above formula result as the combining weights of two kinds of algorithms.According to the sampling statistics of known homogeneous and heterogeneous areas
Data, can convenient select logic function parameter, i.e. midpoint λ0And slope, to ensure required selectivity and flatness.
Virtual more viewization processing
The size for increasing the window of calculating A/G ratio can reduce the variance of estimated value, but will lead to estimation mistake.This is
Because after increasing window, the pixel into window has heterogeneity.In order to solve this problem, we are from non local denoising
Method in take a hint.In order to utilize block similarity measure using a relatively large region of search comprising more pixels
Those are selected to be more likely to and the consistent pixel of object pixel.Specifically, we utilize block in a biggish search range
Similarity measure is found and K most like sub-block of sub-block to be processed.This K sub-block constitutes three-dimensional storehouse.Then to three-dimensional heap
Stack is averaged along the third dimension, calculates A/G statistics by obtained average sub-block.This is equivalent to improving quality of data side
Face has used viewization step more than one, although this is the strategy of a rather rough, has in terms of reliability significant
It improves.
Multi-resolution hierarchy
By establishing three-dimensional storehouse in step (1-2-1), it is similar to virtual visualization and handles, improve weight mapping graph
Quality.Relatively small estimation window is used, reliable statistical result can be also provided, to ensure that high-resolution.However,
If will lead to window edge using the sub-block window of fixed size and occur discontinuously.This phenomenon is since each pixel belongs to
In the estimation window of its many adjacent pixels, their heterogeneity index is affected.For example, the very noisy on homogeneous background
The pixel that will lead in entire window is marked as heterogeneous areas.Finally, the pixel that should belong to homogenous area of neighborhood is worked as
It is denoised as heterogeneous areas, generates artificial artifact in the output.In order to solve this problem, we use one and simply open
Hairdo strategy, i.e. weight of the calculating based on sub-block size, wherein may include biggish weight and lesser weight, then to not
Sub-block with size is averaged.Some unreasonable estimations can be reduced by doing so, and avoid the appearance for manually filtering artifact.
The specific steps of the step 2 are as follows:
Step (2-1): selection LPG-PCA and two kinds of BM3D current state-of-the-art denoising tools.
Step (2-2): LPG-PCA algorithm has stronger inhibition noise immune in the homogenous area of image, therefore in homogeneous
Result after region denoises LPG-PCA assigns biggish weight.
Step (2-3): BM3D is in the details and texture that can protect image well, therefore in the heterogeneous areas pair of image
Result after BM3D denoising assigns biggish weight.
The specific steps of the step 3 are as follows:
Step (3-1): the weight of each sub-block can be determined according to the image weights mapping graph that step (1) obtains, and will
Its output for being used for two filters of linear combination.
Step (3-2): according to step (2), BM3D and LPG-PCA is respectively adopted to noise image and is denoised to obtain two width
Export image；
Step (3-3): using the weight figure calculated, the two width output image that step (2) obtains effectively is merged, right
Homogenous area, the corresponding sub-block weight after the denoising of LPG-PCA is larger, the corresponding son to heterogeneous areas, after BM3D denoising
The weight of block is larger.Image after two methods denoising is effectively merged, and denoising image to the end is obtained.
Although the denoising effect of NLM algorithm is relatively preferable, but still the structure letter for protecting original image that cannot be sufficient
Breath, the image after BM3D algorithm process have a stronger similitude between image block, this method be NLM algorithm performance into
One step improves, and has good protection to image detail, not only has higher signal-to-noise ratio, but also visual effect is more preferable；LPG-PCA exists
Homogenous area noise removal capability is stronger, denoises image clearly, and the image presented after K-SVD processing is more fuzzy, either thin
Section protection or denoising performance are all weaker than LPG-PCA algorithm.
As the reference of fusion for classification algorithms selection, we select BM3D algorithm and LPG-PCA algorithm combination, NLM and K-
Svd algorithm combination, and carry out the comparison of simulation performance.The experimental results showed that BM3D algorithm can have with LPG-PCA algorithm combination
Effect removal noise, and the details and texture of image can be protected well, while avoiding artificial artifact, reach image denoising
Purpose.
Fig. 2 (a)-Fig. 2 (h) is to choose the comparison diagram that different windows carries out soft classification to Lena image.Wherein Fig. 2 (a)
It is original Lena image；Fig. 2 (b) is the weight mapping graph obtained to original Lena image according to 3 × 3 piecemeals；Fig. 2 (c) is pair
The weight mapping graph that original Lena image is obtained according to 5 × 5 piecemeals；Fig. 2 (d) is to obtain to original Lena image according to 7 × 7 piecemeals
The weight mapping graph arrived；Fig. 2 (e) is the weight mapping graph obtained to original Lena image according to 9 × 9 piecemeals；Fig. 2 (f) is pair
The weight mapping graph that original Lena image is obtained according to 10 × 10 piecemeals；Fig. 2 (g) is to original Lena image according to 11 × 11 points
The weight mapping graph that block obtains；Fig. 2 (h) is the weight mapping graph obtained after being averaged to multiple mapping graphs.Obvious Fig. 2 (h) has
Effect avoids artificial artifact.
Fig. 3 (a)-Fig. 3 (h) is the comparison diagram that distinct methods denoising is used to noise-containing Lena image.Wherein, Fig. 3
It (a) is noise image Fig. 2 (a) original Lena image being added after the Gaussian noise that variance is 10；Fig. 3 (b) is to Fig. 3 (a)
Using the image after the denoising of BM3D method；Fig. 3 (c) is to Fig. 3 (a) using the image after the denoising of LPG-PCA method；Fig. 3 (d) is
To Fig. 3 (a) using the image of the fusion after the denoising of BM3D and LPG-PCA method；Fig. 3 (e) is the weight mapping after more views are average
Figure；Fig. 3 (f) is to Fig. 3 (a) using the image after the denoising of NLM method；Fig. 3 (g) is to be denoised to Fig. 3 (a) using K-SVD method
Image afterwards；Fig. 3 (h) is the image to Fig. 3 (a) using the fusion after the denoising of NLM and K-SVD method.Obviously, Fig. 3 (d) vision
Effect is best.
The foregoing is merely preferred embodiment of the present application, are not intended to limit this application, for the skill of this field
For art personnel, various changes and changes are possible in this application.Within the spirit and principles of this application, made any to repair
Change, equivalent replacement, improvement etc., should be included within the scope of protection of this application.
Claims (6)
1. the image de-noising method merged based on homogeneous and heterogeneous areas, characterized in that include:
Step (1): setting the sliding step of window, and original image is divided into several sub-blocks according to being sized for window,
Calculate the weight coefficient of each sub-block；
Step (2): all sub-blocks divided to original image carry out denoising using LPG-PCA algorithm；
To all sub-blocks that original image divides, denoising is carried out using three-dimensional Block- matching BM3D algorithm；
Step (3): according to weight coefficient compared with given threshold, each sub-block is classified as homogenous area or heterogeneous areas；
According to the area classification and weight coefficient of each sub-block, the sub-block after two kinds of algorithms are denoised carries out corresponding fusion, is merged
Image afterwards is final image.
2. the image de-noising method merged as described in claim 1 based on homogeneous and heterogeneous areas, characterized in that the homogeneous
Region refers to: the gray value of all pixels point is in a setting range in the region；The heterogeneous areas refers to: in addition to equal
Region other than matter region.
3. the image de-noising method merged as described in claim 1 based on homogeneous and heterogeneous areas, characterized in that the step
(1) in, the step of calculating the weight coefficient of each sub-block are as follows:
Step (101): the Generalized Likelihood Ratio λ of j-th of sub-block is calculatedj(x) are as follows:
Wherein, G expression is the geometric mean of j-th of sub-block, and A expression is the arithmetic equal value of j-th of sub-block,
N indicates the total number of pixel in j-th of sub-block, xiIndicate ith pixel
Pixel value；
Step (102): according to Generalized Likelihood Ratio λj(x) weights omega (λ of j-th of sub-block is calculatedj):
Wherein, parameter lambda0The median of all sub-block Generalized Likelihood Ratio λ (x) is taken, slope is setting value.
4. the image de-noising method merged as described in claim 1 based on homogeneous and heterogeneous areas, characterized in that the step
(3) the step of are as follows:
To j-th of sub-block of image, first determine whether that it belongs to homogenous area or heterogeneous areas；
If weight coefficient ω (λj) being more than or equal to 0.5, then sub-block belongs to heterogeneous areas；It, will for belonging to the sub-block of heterogeneous areas
The grey scale pixel value of j-th of sub-block after denoising is carried out multiplied by weights omega (λ using three-dimensional Block- matching BM3D algorithmj) after,
Grey scale pixel value with j-th of sub-block for carrying out denoising using LPG-PCA algorithm is multiplied by (1- ω (λj)) sum afterwards, it obtains
The grey scale pixel value of j-th of sub-block of fused image；
If weight coefficient ω (λj) less than 0.5, then sub-block belongs to homogenous area；For belonging to the sub-block of homogenous area, will use
Three-dimensional Block- matching BM3D algorithm carries out the grey scale pixel value of j-th of sub-block of denoising multiplied by weights omega (λj) after, with use
LPG-PCA algorithm carries out the grey scale pixel value of j-th of sub-block of denoising multiplied by (1- ω (λj)) sum afterwards, after obtaining fusion
Image j-th of sub-block grey scale pixel value；
Obtain the grey scale pixel value of all sub-blocks of fused image in turn to get the image after original image denoising is arrived.
5. the image denoising system merged based on homogeneous and heterogeneous areas, characterized in that include: memory, processor and deposit
The computer instruction run on a memory and on a processor is stored up, when the computer instruction is run by processor, in completion
State step described in any one of claim 1-4 method.
6. a kind of computer readable storage medium, characterized in that be stored thereon with computer instruction, the computer instruction is located
When managing device operation, step described in any one of the claims 1-4 method is completed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811061739.8A CN109242797B (en) | 2018-09-12 | 2018-09-12 | Image denoising method, system and medium based on homogeneous and heterogeneous region fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811061739.8A CN109242797B (en) | 2018-09-12 | 2018-09-12 | Image denoising method, system and medium based on homogeneous and heterogeneous region fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109242797A true CN109242797A (en) | 2019-01-18 |
| CN109242797B CN109242797B (en) | 2021-10-19 |
Family
ID=65067611
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811061739.8A Expired - Fee Related CN109242797B (en) | 2018-09-12 | 2018-09-12 | Image denoising method, system and medium based on homogeneous and heterogeneous region fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109242797B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111612706A (en) * | 2020-04-28 | 2020-09-01 | 广州科易光电技术有限公司 | Filtering method and system applied to infrared image |
| US20220253985A1 (en) * | 2019-10-31 | 2022-08-11 | Shenzhen Institutes Of Advanced Technology | Method for denoising image, apparatus, and computer-readable storage medium |
| US20230130835A1 (en) * | 2021-10-25 | 2023-04-27 | Realtek Semiconductor Corp. | Image processing system and related image processing method for image enhancement based on region control and texture synthesis |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120281923A1 (en) * | 2011-05-02 | 2012-11-08 | Yeda Research And Development Co., Ltd. | Device, system, and method of image processing utilizing non-uniform image patch recurrence |
| CN103093434A (en) * | 2013-01-27 | 2013-05-08 | 西安电子科技大学 | Non-local wiener filtering image denoising method based on singular value decomposition |
-
2018
- 2018-09-12 CN CN201811061739.8A patent/CN109242797B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120281923A1 (en) * | 2011-05-02 | 2012-11-08 | Yeda Research And Development Co., Ltd. | Device, system, and method of image processing utilizing non-uniform image patch recurrence |
| CN103093434A (en) * | 2013-01-27 | 2013-05-08 | 西安电子科技大学 | Non-local wiener filtering image denoising method based on singular value decomposition |
Non-Patent Citations (2)
| Title |
|---|
| CHARLES-ALBAN DELEDALLE等: "《Iterative weighted maximum likelihood denoising with probabilistic patch-based weights》", 《 IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
| 侯建华 等: "《基于广义似然比的小波域SAR图像相干斑抑制算法》", 《中南民族大学学报(自然科学版)》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220253985A1 (en) * | 2019-10-31 | 2022-08-11 | Shenzhen Institutes Of Advanced Technology | Method for denoising image, apparatus, and computer-readable storage medium |
| CN111612706A (en) * | 2020-04-28 | 2020-09-01 | 广州科易光电技术有限公司 | Filtering method and system applied to infrared image |
| CN111612706B (en) * | 2020-04-28 | 2023-10-13 | 广州科易光电技术有限公司 | Filtering method and system applied to infrared image |
| US20230130835A1 (en) * | 2021-10-25 | 2023-04-27 | Realtek Semiconductor Corp. | Image processing system and related image processing method for image enhancement based on region control and texture synthesis |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109242797B (en) | 2021-10-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | Noise detection and image denoising based on fractional calculus | |
| CN108921800B (en) | Non-local mean denoising method based on shape self-adaptive search window | |
| CN103236046B (en) | Based on the fractional order adaptive coherent spot filtering method of image aspects fuzzy membership | |
| Zhong et al. | Image denoising based on wavelets and multifractals for singularity detection | |
| Ma et al. | An improved Sobel algorithm based on median filter | |
| CN110443775B (en) | Discrete wavelet transform domain multi-focus image fusion method based on convolutional neural network | |
| CN110675340A (en) | Single image defogging method and medium based on improved non-local prior | |
| Zhang et al. | Decision-based non-local means filter for removing impulse noise from digital images | |
| CN107909560A (en) | A kind of multi-focus image fusing method and system based on SiR | |
| CN109509163B (en) | FGF-based multi-focus image fusion method and system | |
| CN109242797A (en) | Image de-noising method, system and the medium merged based on homogeneous and heterogeneous areas | |
| Deng et al. | Speckle reduction of ultrasound images based on Rayleigh-trimmed anisotropic diffusion filter | |
| Raj et al. | Medical image denoising using multi-resolution transforms | |
| Alenezi et al. | Geometric-pixel guided single-pass convolution neural network with graph cut for image dehazing | |
| CN104657951A (en) | Multiplicative noise removal method for image | |
| CN110796616A (en) | Fractional order differential operator based L0Norm constraint and adaptive weighted gradient turbulence degradation image recovery method | |
| Joel et al. | Nonsubsampled contourlet transform with cross‐guided bilateral filter for despeckling of medical ultrasound images | |
| Li et al. | Extended neighborhood-based road and median filter for impulse noise removal from depth map | |
| CN111461999A (en) | SAR image speckle suppression method based on super-pixel similarity measurement | |
| Rama Lakshmi et al. | A Review on Image Denoising Algorithms for Various Applications | |
| CN110852975A (en) | Method for jointly driving non-local mean value of radar image by using variation coefficient and local entropy | |
| CN105096274A (en) | Infrared image noise reduction method based on non-subsampled contourlet domain mixed statistical model | |
| CN112734666B (en) | SAR image speckle non-local mean suppression method based on similarity value | |
| Shankar et al. | Object oriented fuzzy filter for noise reduction of Pgm images | |
| Yang et al. | Hierarchical joint bilateral filtering for depth post-processing |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20211019 |